Fiducial marks is a set of marks located in the corners or edge-centers, or both, of an aerial photographic image. These marks are exposed within the camera onto the original film and are used to define the frame of reference for spatial measurements on aerial photographs. Opposite fiducial marks connected, intersect at approximately the image center or principal point of the aerial photograph. The principal point is the geometric center of the photograph. Typical positions of fiducial marks in an aerial photo are shown in Figure 1, and some example shapes of fiducial marks are given in Figure 2.
Interior orientation is concerned with the determination of a set of parameters for the transformation from pixel to image coordinates. For digital cameras the relationship between pixel and image coordinates is constant and is determined during the calibration stage. If film images are scanned in a separate step (which is the case in digital photogrammetry today) the pixel-image relationship must be established for each digital image individually . Interior orientation involves the process of measuring the fiducial marks on the digital imagery. These can be automated in two ways: by performing correlation or by image analysis using certain knowledge about the fiducials . For the correlation approach, a template image has to be made representing the ideal image of a fiducial mark (a replica of those present in real images). Then this template is matched with the scanned image containing the fiducial marks. The second approach is based on image analysis. For this analysis, prior knowledge of the geometry of the fiducial mark is necessary, and this knowledge helps to design an automated procedure to segment fiducial marks in an aerial image.
Image processing and computer vision techniques have successfully been employed for facilitating automated procedures in digital aerial images such as interior orientation. Lue  introduced a fully automatic digital interior orientation based on the template matching techniques using a database containing fiducials of widely distributed aerial cameras. Schickler and Poth  presented a binary cross correlation method in an image pyramid. Kersten and Haering  described a fully operational automatic interior orientation for digital aerial images based on a modified Hough Transform for rough localization of fiducial marks and Least Squares Matching for precise measurement. The main disadvantage of current least squares matching for fiducial marks is: 1) a database is necessary and 2) they need to know roughly where the fiducial mark is before least squares matching can be performed (the approximated fiducial position cannot be far away from the true position; only a few pixels are allowed otherwise least squares matching will get wrong answers).
In this paper we will use image analysis approach to obtain the positions of fiducial marks in an aerial photograph. We will use advanced image analysis technique such as the attribute-based morphology to initially segment out the fiducials automatically, and then refine the positions to achieve sub-pixel accuracy.
The advantage of our paper is to automatically locate and segment a fiducial mark from a large area (if only one fiducial is occurred in the area, sub-image of the corners or middle side of the input image). Once the approximated locations and segmentations are determinated, we can perform the extraction of centers of fiducials precisely using: 1). Location operators: which does not need a fiducial database; only the shape property of fiducials is used in our paper. It is not state-of-the-art. However, it is very practical because a database is not always avaliable to users. 2). Least squares matching if a database is established (our further work).
Figure 1: Illustration of the location of the fiducial marks and principal point for an aerial photo image. The intersection of the dotted line indicates the principle point.
Figure 2: Illustration of several typical shapes of fiducial mark in an aerial photo image.
Section 2 shows the methods for fiducial marks segmentation. The refinement of the positional information is given in Section 3. Section 4 shows the results obtained using our automatic segmentation procedure. Section 5 concludes.